Our main goal is to improve protein structure prediction methods in order to develop models of proteins in biologically relevant states. Such states may include the target protein as a homo-oligomer;complexed with other proteins, nucleic acids, and ligands;covalently modified through phosphorylation and glycosylation;and in alternate physiologically relevant conformations. Information on the structure of these states for any one target may come from a number of different templates;this information can be assembled into a composite model from which biological inferences can be made. The next generation of the backbone-dependent rotamer library will be developed using classical and Bayesian non-parametric statistics, and it will be extended to include protein modifications, such as phosphorylated and glycosylated amino acids. Electron density analysis will be used to exclude residues with uncertain or dynamic conformations. The resulting libraries will be incorporated into the next generation of our widely used side-chain prediction program SCWRL. A very general structural bioinformatics platform will be constructed to enable statistical and conformational analysis of protein structures on a routine basis. We propose to develop interactive methods and software for producing biologically meaningful models of proteins and protein complexes, based on multiple structure alignments, hidden Markov models, and combined information from diverse structures - ligand-bound and unbound structures, monomers and homo-oligomers, and protein complexes. Project narrative Knowledge of protein structures and their complexes is vital to understanding function and mechanism. We will develop algorithms, databases, and software for predicting structure in biologically relevant states, including homo-oligomers, post-translational modifications, and protein complexes. These methods will be used to improve human health through the prediction of proteins involved in disease.